Interesting! At 0:12 there is a frame where two balls are close and the tracking didn't work. Is the reason the code sought something more like a ball and found an eight?
Hey there! Good spot! So basically there’s two stages to object tracking, the actual object detection, and then assignment. This is a demonstration of object detection, where the task is simply to extract the likely ball positions in the frame. If there is partial occlusion (like at that moment you pointed out), the algorithm sometimes treats the detection as a single object. This is not an issue as during the assignment phase, a kalman like filter algorithm predicts the position of the partially hidden object, so 9 objects are always accounted for. Let me know if you have any specific questions 🙂
@@jamescozensjuggling6868 what is the delay between the moment the ball is in the position N and the moment you can get the data of this position? With such thing skynet will never miss a shot ☠️☠️☠️